Nodes/Comfyui-PixNodes/Qwen API Assistant (PixNodes)
ComfyUI Node

Qwen API Assistant (PixNodes)

Alibaba's Qwen models as a graph node — including vision and OCR

By pixixai·Created 8 months ago·Updated 6 months ago· 24
Qwen API Assistant (PixNodes)
  • images
  • text_response
  • history
api_key
base_urlhttps://dashscope.aliyuncs.com/compatible-mode/v1
modelqwen-plus
prompt
system_instructionYou are a helpful assistant.
seed0
max_tokensauto
temperatureauto
top_pauto
history
input_file_text

Qwen is Alibaba's model family, and this node puts the cloud versions of it in your graph as a chat client: prompt in, text out, nothing runs locally. It's the Qwen sibling of the pack's Kimi API Assistant - same shape, same key-saving trick, different provider. If you're on Alibaba Cloud's DashScope, or you want a strong Chinese-market LLM (and its vision/OCR models) wired into your workflows, this is the one.

It speaks OpenAI's chat-completions dialect against base_url, which defaults to DashScope's compatible endpoint: https://dashscope.aliyuncs.com/compatible-mode/v1. Your API key comes from the DashScope console, not from anywhere in the pack.

The model dropdown

Six choices, and the split is worth knowing:

  • Text: qwen-max, qwen-plus, qwen-flash - quality ladder, roughly. Plus for everyday, Flash for cheap-and-fast, Max when you want the best.
  • Vision: qwen-vl-max, qwen-vl-plus - these accept an image.
  • OCR: qwen-vl-ocr - text extraction from images, which is a genuinely useful niche (read a screenshot, a sign, a panel).

The default is qwen-plus, which is a text model - so if you're about to feed it an image, switch to a -vl- variant first, or you'll get an error back.

The inputs

api_key (saved after first use to ComfyUI/user/PixNodes/api_key.json), prompt, system_instruction, seed, and max_tokens/temperature/top_p all defaulting to "auto" (skipped when "auto"). Optional: history (multi-turn via the history output), input_file_text (appended as [File Context]), and images (IMAGE - encoded to base64 PNG and sent to the vision models).

Outputs: text_response (STRING) and history (STRING, the full message list to feed back in for follow-ups).

Install

Part of Comfyui-PixNodes. ComfyUI Manager → search "PixNodes" → install → restart, or:

cd ComfyUI/custom_nodes
git clone https://github.com/pixixai/Comfyui-PixNodes

then restart. (The README's manual-install line pastes the wrong repo URL, ComfyUI-AlignLayout - use the one above.) Dependencies are openai + Requests; if Manager didn't install them, pip install openai from the pack folder.

The honest warnings

Same three cautions as its Kimi sibling, so I'll keep them short: errors come back as Error:-prefixed strings, not node failures; the saved key is plaintext JSON on your disk; and it's a metered, networked API call - the security frame for API-wrapper nodes applies, since "node that phones home with your key" is exactly the shape that's been abused in this ecosystem before. Read the requests if you're paranoid, and prefer this node for jobs a cloud LLM is genuinely good at: prompt enhancement, captioning a frame with a vision model, or OCR'ing a sign. For "run a small local model offline instead," this isn't it - that's a different category of node entirely.

CategoryPixNodes/LLM

Inputs (12)

NameTypeDefaultDescription
api_keySTRING
base_urlSTRINGhttps://dashscope.aliyuncs.com/compatible-mode/v1
modelCOMBOqwen-plus6 options: qwen-max, qwen-plus, qwen-flash, qwen-vl-max, qwen-vl-plus, qwen-vl-ocr
promptSTRING
system_instructionSTRINGYou are a helpful assistant.
seedINT00–2147483647
max_tokensSTRINGauto
temperatureSTRINGauto
top_poptSTRINGauto
historyoptSTRING
input_file_textoptSTRING
imagesoptIMAGE

Outputs (2)

NameTypeDescription
text_responseSTRING
historySTRING